1

Bribery environments and firm performance: Evidence fromCEE countries

Jan Hanousek†

Anna Kochanova‡

Abstract

We examine the relation between bureaucratic corruption and firm performance in CEEcountries. We show that divergent consequences of corruption found in previous studies canbe explained by the specifics of the local bribery environment in which firms operate. Ahigher mean bribery is associated with lower performance, while higher dispersion ofindividual firm bribes appears to facilitate firm performance. We also conduct a detailedanalysis by firm sector and size, and countries’ institutional environments.

Keywords: bureaucratic corruption, firm bribing behavior, firm performance, CEE countries.

JEL classifications: D22, D73, O12, P37.

ACKNOWLEDGEMENTS: We thank Alena Bicakova, Marina Dodlova, Randall Filer, Vahagn Jerbashian,Stepan Jurajda, Evzen Kocenda, and Patrick Warren for valuable comments. An earlier draft of this paperbenefitted from comments from participants at the Economic Governance and Innovation Conference 2011,ESNIE 2012, and various seminars. Anna Kochanova is grateful to the financial support of GDN grant No.RRC 11-004 and Czech Science Foundation project No. P402/12/G097 DYME. Jan Hanousek is grateful tosupport from GACR grant No. 14-31783S. The usual disclaimer applies.† CERGE-EI, Politickych veznu 7, 11121 Prague, Czech Republic; CEPR, London. E-mail:jan.hanousek@cerge-ei.cz. CERGE-EI, a joint workplace of the Center for Economic Research andGraduate Education, Charles University in Prague and the Economic Institute of Academy of Sciences ofthe Czech Republic.‡ Corresponding author: Max Planck Institute for Research on Collective Goods. Kurt-Schumacher-Str. 10,53113 Bonn, Germany. Email: kochanova@coll.mpg.de.

2

1. Introduction

In countries with weak policies and legal systems corruption is considered a strong

constraint on growth and development. The existing literature on the effects of corruption

on firm performance is, however, divided. One branch considers corruption a ‘grease the

wheels’ instrument that helps overcome cumbersome bureaucratic constraints, inefficient

provision of public services, and rigid laws (Huntington, 1968; Lui, 1985; Lein, 1986),

especially when countries’ institutions are weak and function poorly (Acemoglu and

Verdier, 2000; Meon and Weill, 2010; De Vaal and Ebben, 2011). Another branch argues

that corruption reduces economic performance due to rent seeking, increase of transaction

costs and uncertainty, inefficient investments, and misallocation of production factors

(Murphy et al., 1991; Shleifer and Vishny, 1993; Rose-Ackerman, 1997; Kaufmann and

Wei, 2000). Empirical evidence at the firm-level is also ambiguous (McArthur and Teal,

2004; Fisman and Svensson, 2007; De Rosa et al., 2015; Vial and Hanoteau, 2010), and

overall remains scarce due to the lack of available data.

In this paper we contribute to the firm-level empirical research on bureaucratic

corruption and firm performance, and explain the divergent effects of corruption found in

previous studies. We employ a rich firm-level panel dataset with widely accepted measure

of bureaucratic corruption (bribery) that allows us to alleviate some of the methodological

concerns of existing research. We focus on a group of countries from CEE, as they have

similar history of transition to market economy, but are still institutionally diverse.

Our approach is to combine the information on firm bribery practices from BEEPS

and firm financial data from the Amadeus database. This gives us a large firm-level panel

data for 14 CEE countries over 1999 – 2007, which have more accurate and detailed

information on firms’ economic activity and bribery than BEEPS alone. Previous studies

that use firm bribery practices and performance from anonymous surveys such as BEEPS or

WBES suffer from missing data, as firms often reluctant to reveal their financial records

3

(Gaviria, 2002; McArthur and Teal, 2004; Fisman and Svensson 2007; De Rosa et al.,

2015).1 Those studies also deal with cross-sectional data, while we are able to explore the

panel structure of our dataset. In the regression analysis we control for firm fixed effects,

which eliminate time-invariant factors that could simultaneously cause bribery and firm

performance. This is an important step to diminish the endogeneity of bribery measure,

given the recognized difficulties to find exogenous variation to explain corruption.

To combine two datasets we introduce ‘local bribery environments.’ We define

‘local markets,’ in which firms operate, as clusters formed by survey wave, country, double-

digit industry, firm size, and location size. This is relevant, since bureaucratic corruption

might be a local phenomenon that depends on not only on country, but industry, firm and

markets size. Within local markets we analyze how the ‘local bribery environments’ –

characterized by the means and dispersions of individual firm bribes – influence the

economic performance of firms. We compute the mean and standard deviation of the

bribery measure from BEEPS for the universe of local markets. For firms from the Amadeus

we can also identify those markets, and thereby each firm is assigned characteristics of local

bribery environment. Economically, the mean bribery approximates the equilibrium level of

bribery in a local market. The bribery dispersion, meanwhile, represents the pervasiveness

of bureaucratic corruption and availability of opportunities to extract benefits from bribery

for some firms.

The use of the notion of local bribery environments is another step to reduce the

endogeneity of bribery measures, since an individual firm performance less likely affects the

bribery environment, than its own bribing behavior. The joint use of two independent data

sources also alleviates this concern, because not the same firms report financial statements

1 For example, in the widely-used BEEPS and WBES (databases about 40 to 50% of firms do not report theirperformance indicators. BEEPS ( Business Environment and Enterprise Performance Survey) is a part ofthe global WBES (World Bank Enterprise Survey).

4

and bribery. Further attempts to deal with endogeneity would likely reduce the endogeneity

bias in the same direction, therefore, our empirical results are estimated at the lower bound.

Similar to many papers (Gaviria, 2002; Beck et al., 2005; Fisman and Svensson,

2007; Vial and Hanoteau, 2010) we measure firms performance as sales and labor

productivity growth of firms, as these enhance wealth and employment creation, and

stimulate economic development. The results of the empirical analysis, identified from

within-firm variation, show that the ambiguous consequences of corruption found in

previous studies can be explained by the different effects of the mean and dispersion of

bureaucratic corruption in the local environment. In particular, a higher bribery mean

impedes both the real sales and the labor productivity growth of firms. This is generally

consistent with the existing firm- and macro-level empirical research. In contrast, a higher

dispersion of individual firm bribes facilitates firm performance. Moreover, firms are more

likely increase their growth rates in the environments with higher both bribery mean and

dispersion. We also find that these impacts are more pronounced in the case of labor

productivity growth. These results are robust to various specification checks.

Our results suggest that in more dispersed local bribery environments at least some

firms that bribe receive preferential treatments from public officials, and non-bribing firms

are likely to be more efficient in production and growth. The existence of a certain number

of bribing firms in a local market, therefore, stimulates aggregate firm performance. This

finding is in line with Acemoglu and Verdier (2000), as positive effects from bribery

dispersion can overshoot negative effects from bribery mean. The chance to receive benefits

from bribery for particular firms may be one reason why corruption does not vanish in spite

of its overall growth restraining effect (Mauro, 1995). In addition, we find that our results

vary for different types of firms. Smaller and more stable firms are least affected by bribery,

while service firms are able to gain most in environments with higher corruption dispersion.

5

We also observe that in countries with stronger institutions, the effects of bribery mean and

dispersion are more pronounced.

The reminder of the paper is structured as follows. Section 2 introduces the notion of

‘local bribery environments’ and discusses its relation to firm performance. Section 3

describes the data and explains the merging of the financial information and the bribery

practices of firms. Section 4 outlines the empirical methodology. Section 5 presents the

results and robustness checks, and section 6 concludes.

2. Local Bribery Environments and Firm Performance

The institutional environment of a country largely determines its level of economic

development, overall corruption, and the behavior and performance of firms (Acemoglu,

2003). However, a country may consist of many narrow local markets that can be

heterogeneous with respect to economic conditions as well as bribery practices. A small

furniture company located in a rural area, for instance, may face a different demand for and

provide a different supply of bribes than a large retail firm located in a capital city.

In this paper we focus on local markets that are comprised of firms sharing a similar

size, area of economic activity (industry), and location size. These local markets we

characterize by the levels of bribery mean and dispersion of individual firm bribes, which

we term the ‘local bribery environments.’2 Bribery mean can be viewed as an equilibrium

level of corruption in a local market, defined by the demand from public officials and

supply by firms. Bribery dispersion reflects firms’ willingness to bribe (Bliss and Tella,

1997; Svensson, 2003; Luo and Han, 2008), the discretionary power of public officials and

uncertainty in a local market. Ceteris paribus, higher bribery dispersion suggests lower

2 The notion of the ‘local bribery environment’ is aligned with the arguments of Svensson (2003) and Fismanand Svensson (2007) that bribery is industry- and region-specific. They suggest that a firm depends more onpublic officials, and therefore might have to pay higher bribes (or pay bribes more often) if it requires morepermits and licenses due to specifics of its economic activity or location. However, it is unlikely that all firmsin a local market always bribe equally.

6

pervasiveness of bureaucratic corruption but higher opportunity to extract benefits from

bribes for some firms.

Higher bribes can alter firms’ incentives to grow, such that they prefer to remain

small and less visible to public officials (Gauthier and Goyette, 2014). Bribery can also

restrain firms from obtaining licenses and permissions, which undermines innovations and

investment (O’Toole and Tarp, 2014), and can limit exporting and importing activities

essential for firm growth. It can also cause longer delays in public services provision and

thereby project interruptions if bureaucrats tend to increase red tape in order to extract more

bribes (Kaufmann and Wei, 2000). Finally, higher bribes can provoke reallocation of talent

from production to rent-seeking (Murphy et al., 1991, Dal Bo and Rossi, 2007). In this case,

one would expect a negative relationship between bribery mean and firm performance.

Some empirical research finds either an insignificant or negative impact of bribery on the

sales growth or productivity of firms (for example, Gaviria, 2002; McArthur and Teal,

2004; Fisman and Svensson, 2007). For CEE and the former Soviet Union countries, De

Rosa et al. (2015) show that bribery more negatively affects firm productivity in non-EU

countries and in those with weaker institutions.

However, if bribery works as a ‘grease the wheels’ instrument, it can help overcome

some bureaucratic constraints and inefficient public services provision (Huntington, 1968;

Lui, 1985; Lein, 1986; and Vial and Hanoteau’s, 2010 presents empirical evidence for

Indonesia). That would create a positive relationship between bribery mean and firm

performance.

Given a positive level of bribery mean, in an environment with low bribery

dispersion all firms bribe in the same way. Corruption is pervasive and can be seen as a tax

or an additional fee for public services provision. This would not create distortions other

than those connected to the bribery mean.

7

In an environment with higher bribery dispersion, only part of firms bribes. If

bribery benefits all or the majority of bribing firms a higher dispersion can facilitate joint

firm performance. This situation could happen when bribing firms are able to exploit

favorable opportunities from bribery, and are efficient in giving bribes, while public

officials are able well discriminate firms to extract bribes in a local market. Non-bribing

firms must be efficient in complying with bureaucratic regulations, and benefit from better

allocation of their production recourses, as otherwise, bribing firms would crowd out those

that do not bribe.3

However, if bribery helps only a minority of bribing firms, creates negative

externalities (Kaufmann and Wei, 2000), and does not incentivize non-bribing firms to

perform better, then in a more dispersed local bribery environment firm performance can

deteriorate. Finally, higher bribery dispersion can be perceived as a higher uncertainty in a

local market. If all firms bribe occasionally and public officials are unpredictable, higher

dispersion would be also negatively associated with firm performance.

In this paper we aim to analyze the relationship between the characteristics of the

local bribery environment and firm performance.4 In the next section we describe our data

and the definition of local bribery environments for the empirical analysis.

3. Data

3.1. Data Sources

The bribery measure is taken from BEEPS, an anonymous survey of a stratified random

sample of firms from CEE and former Soviet Union countries.5 It consists of a rich set of

3 Hanousek and Palda (2009), for example, show that in an uneven environment, efficient non-tax-evadingfirms are crowded out by inefficient tax-evading firms.4 The analysis of the specific channels through which bribery can impact firm growth, however, is beyond thescope of this paper.5 BEEPS is collected jointly by the World Bank and the European Bank for Reconstruction and Development.The data are available online at https://www.enterprisesurveys.org and at http://ebrd-beeps.com/data/. Datafor this paper was downloaded from the first source.

8

questions about firms’ activity, market orientation, financial performance, and employment

as well as infrastructural, criminal, corruption, and legal environments. A disadvantage of

BEEPS is missing data for questions related to accounting information (40–50% missing

data on sales, assets, costs, etc.), which can imply a biased inference from the data analysis.

For instance, the worst-performing firms may not report their accounting information and

complain more about corruption (Jensen et al., 2010). Each wave of BEEPS covers the three

preceding years; we use the three waves completed in 2002, 2005, and 2008.

The financial data comes from the Amadeus database. It contains detailed balance

sheet and income statement data, industry codes as well as the exact identification of

European firms.6 Because non-active (unresponsive or exiting from the market) firms are

excluded from the database after a certain period, we have merged several editions of

Amadeus (2003, 2007, and 2010).

For the analysis we chose 14 CEE countries that are well covered in both Amadeus

and BEEPS: Bulgaria, Croatia, the Czech Republic, Estonia, Hungary, Latvia, Lithuania,

Poland, Romania, Russia, Serbia, Slovakia, Slovenia, and Ukraine. These countries are

similar in that they started the transition to a market economy at approximately the same

time. They are, however, quite different in overall corruption levels, as Figure 1 in

Appendix shows for the Control of Corruption indicator from the Worldwide Governance

Indicators (WGI) database compiled by the World Bank.

Both the BEEPS and Amadeus databases tend to understate very small firms, and

Amadeus tends to overstate large firms (Klapper et al., 2006).7 Therefore, we conduct our

analysis for different subsamples of firms, in particular, for firms of different sizes and

industrial sectors.

6 Details about the Amadeus database can be found at http://www.bvdep.com.7 A detailed comparison of eight OECD countries with the whole population of firms retrieved fromOECD.STAN is available in the Online Appendix.

9

3.2. Combining Information from the BEEPS and Amadeus Databases

The joint use of the BEEPS and Amadeus databases provides a good opportunity to study

the effects of local bribery environments on firm performance. To combine bribery practices

with firm financial information we define clusters that represent local markets, using the

following criteria: i. country; ii. time period (1999–2001, 2002–2004, and 2005–2007,

corresponding to the waves of BEEPS); iii. industry (two-digit ISIC rev 3.1 industry code),

4) firm size (micro firms with 2–10 employees, small firms with 11–49 employees, and

medium and large firms with more than 50 employees); and iv. location size (capital, city

with population above 1 million, and city with population below 1 million).8

It is straightforward to identify clusters in both databases. In BEEPS and Amadeus

firms report industry and employment data. In BEEPS firms record the size of location. In

Amadeus firms report the address of registration, which we use to identify capitals and cities

with population above 1 million (these are only in Russia and Ukraine) to construct a

location size variable. For each cluster we compute the mean and standard deviation of the

bribery measure from BEEPS and assign them to every firm in the Amadeus database

operating in the same cluster. Given the structure of the data, the mean and standard

deviation of the bribery measure are the way to describe bureaucratic corruption

environment in the local market. They represent an equilibrium bribery level and the

dispersion of individual firm bribes.9

8 We cannot utilize ‘regions’ in the criteria defining local markets, as would be in accordance with Svensson(2003) and Fisman and Svensson (2007), since regions are not consistently defined in BEEPS. In therobustness check, therefore, we show that the results of this study remain the same for the subsample offirms located in the capital cities only and for the case when size of location is omitted from the criteriadefining clusters.9 Anos-Casero and Udomsaph (2009) and Commander and Svejnar (2011) also attempt to combine twodatasets using the 2002 and 2005 waves of BEEPS for 7 – 8 CEE countries. Our main departure from thesepapers is that we separate micro firms with fewer than 10 employees from small firms with 11-49 employees.This is motivated by the fact that originally nearly 45% of firms in BEEPS and 40% of firms in Amadeus aremicro firms. Clearly, micro firms might be exempted from some bureaucratic regulations and taxes (WB,2004; EC, 2011), and consequently they may encounter demands from public officials less often. Anos-Caseroand Udomsaph (2009) and Commander and Svejnar (2011) study how business constrains impact TFP andefficiency to generate revenue. A recent paper by Fungacova et al. (2015) uses exactly the same criteriadefining clusters as we do. It studies whether bribery affects firm-level bank debt. None of these papers,however, examine the dispersion of individual firm bribes or reported business constraints within clusters.

10

The criteria defining clusters explain 40% of the total variation of the bribery

measure in BEEPS.10 We require each cluster to have at least 4 firms, which reduces sample

size to 10,097 firms (67% of the original sample), and we obtain 1,137 cells in BEEPS. The

average number of firms in a cell is 8.87 and the median is 6. Only two clusters computed

using BEEPS have no counterparts in Amadeus. About 48% of observations from Amadeus

are assigned characteristics of the local bribery environments derived from BEEPS,11 which

yields around 700,000 observations.

Our final dataset results in unbalanced panel data for 1999–2007 where the bribery

measures remain constant over three-year periods. Besides availability of high quality firm-

level financial data, the advantage of our dataset is the alleviation of the endogeneity

between firm performance and bribery, which we discuss in detail in the next section.

Another advantage is the reduction of measurement error and firm-specific perceptions (due

to managers’ optimism or pessimism) in the bribery mean measure by averaging them out.

3.3. Definitions of Variables

The bribery measure is obtained from answers to the following BEEPS question:

Thinking about officials, would you say the following statement is always, usually,

frequently, sometimes, seldom, or never true: “It is common for firms in my line of business

to have to pay some irregular “additional payments/gifts” to get things done with regard to

customs, taxes, licenses, regulations, services, etc.?12

10 This result is R2 obtained from the analysis-of-variance (ANOVA) with the bribery measure as adependent variable and all interactions between country, year, industry, firm size, and location size asindependent variables.11 A complete number of clusters in the roster would be 8100=14(country)*3(wave)*2(3 for Russia andUkraine, location size)*3(firm size)*30(industry), but because BEEPS does not cover all firms, industries, etc.combinations, and we disregard cells with less than 4 firms, we have only 1,137 cells (14%). The Amadeusdatabase has the best coverage of the firm financials for the CEE countries, therefore, 48% of observationsfrom Amadeus merged with BEEPS is a large number. Additional summary statistics are available in theOnline Appendix.12 The framing ‘in my line of business’ or ‘typical firm like yours’ is a common approach to provide moreconfidence to respondents and at the same time to elicit their own experience.

11

Amongst the questions about corruption, this one is the most neutral, and virtually

the only one that occurs consistently across all three waves. The variable is categorical and

takes values from 1 to 6. For convenience purposes we rescale it to a variable that varies

from 0 to 1 by subtracting 1 from the original value and dividing the result by 5. The

transformed variable can also be interpreted as the intensity of bribery or the probability to

bribe. Figure 2 in Appendix offers a country-time variation of this measure. It is

heterogeneous across countries and overall decreases with time.

For our performance variables we consider real sales growth and real labor

productivity growth as used in previous studies (Gaviria, 2002; Beck et al., 2005; Fisman

and Svensson, 2007; Vial and Hanoteau, 2010).13 Real sales are approximated by the firm

operational revenue in 2000 prices, and labor productivity is real sales per employee. We

take first differences of the logarithms of these measures to derive growth rates. Further, in

the regression analysis we use three-year averages of these growth rates to match the

variation of bribery mean and dispersion. We expect that a local bribery environment may

have a somewhat different effect on these performance measures. We opt for the analysis of

sales, as company turnover is not directly affected by corporate income taxes and transfers.

On the other hand, labor productivity should reflect changes in employment structure and

therefore reveal firm performance potential in a longer horizon. The dynamics of these firm

characteristics are important for development as they enhance economic welfare and

employment creation.

For controls we employ the usual set of variables used in the firm-level financial

studies. To proxy firm size we use the logarithms of total assets and number of employees,

as well as their squares to control for possible non-linearity. Market share is computed as

the ratio of sales of a firm to total sales in an industry defined at the four-digit level. Firm

13 We do not measure productivity as TFP (total factor productivity) or value-added per employee, becauseAmadeus has many missing values in the intermediate material and staff cost variables for CEE countries;Russia, Latvia, and Lithuania do not report them at all. We use a simplified version of productivity that allowsfirms’ capital and intermediate costs to be flexible.

12

profitability is defined as EBIT (earnings before interest and taxes) over total assets.

Leverage equates to book leverage ratio – total debt over total assets, and cash flow is the

reported cash flow scaled by total assets.

These control variables can correlate with bribery measures and reduce the omitted-

variable bias. Firms with lower market shares, for instance, can be more engaged in bribery

in order to survive on the market. Luo and Han (2008) report such a correlation in a study of

the determinants of bribery and graft using WBES for several developing countries. More

profitable firms may have a higher willingness to pay and can pay larger bribes and/or more

frequently (Bliss and Tella, 1997; Svensson, 2003). Firm leverage can also correlate with

bribery if unofficial payments are needed to obtain external financing (Beck et al., 2005;

Fungacova et al., 2015). The availability of cash can also open greater opportunities for

bribe payments. In addition, the control variables restrict the sample to those firms that

report all essential financial information, making it more homogeneous across countries. We

measure these variables in 1999, 2002 and 2005. Summary statistics of all employed

variables and their pairwise correlations are in Tables 1 and 2.

4. Empirical Methodology

The identification of the relation between bribery and firm performance is not

straightforward due to possible endogeneity. Bribery may influence firm performance by

increasing or reducing constraints on operation and growth, while better performing firms

may have a greater willingness and ability to pay bribes. This reverse causality can be

induced by further unobservable factors that correlate with both firm performance and

bribery practices, such as managerial talent and firm culture.

In the context of this study, the endogeneity problem is largely reduced due to

several facts. First of all, the panel structure of the data allows us to control for firm fixed

effects and to remove time-invariant unobservable factors that could potentially cause both

13

firm performance and bribing behavior. The short length of our panel increases the

likelihood of these factors being fixed over time. Second, instead of bribing behavior of

individual firms, we employ more aggregated measures – bribery mean and dispersion in a

local market defined by industry, firm-size, and location-size characteristics. Arguably, an

individual firm has a negligible influence on these aggregate measures.14 This influence is

further decreased when firm performance and bribery measures come from different

independent data sources (Anos-Casero and Udomsaph, 2009).

Nevertheless, in the next section we first compare the estimates identified from

within-firm variation with the estimates identified from within-cluster variation to

demonstrate the reduction of the endogeneity bias. This occurs because average firm

performance within a cluster more likely affects mean bribery, inducing an upward bias of

the estimates (if better performing firms are ready to bribe more frequently). Admittedly,

firm fixed effects do not account for temporal endogeneity. The bias due to temporal

endogeneity, however, has likely the same direction as the bias due to permanent

endogeneity. Our estimates, therefore, are at the lower bound.

Our empirical specification is a typical growth equation, originally proposed by

Evans (1987), where the dependent variable is the growth rate and the independent variables

are lagged to control for initial conditions.15

yit = β0 + β1Bribery Meanct + β2Bribery Dispersionct + γXit−1 + υi + νt + ςs + εit,

(1)

14 In view of the difficulty to find appropriate instruments for bribery measures, the use of industry-location or industry-location-firm size average measures of bribery or obstacles to firm growth and operationinstead of firm-specific measures is a handy approach to reduce the endogeneity problem in existing research,which employs cross-sectional data from BEEPS, WBES, or IC (Investment Climate). See, for example,Dollar et al. (2005), Aterido et al. (2011), and Commander and Svejnar (2011).15 Similar specifications are also widely used in the literature that studies the effects of privatization, politicalconnections, and other events on firm performance, see, for example, Hanousek et al. (2007) and Boubakri etal. (2008).

14

where yit is the performance measure of firm i at time period t; it is either real sales or labor

productivity growth rates, averaged over three-year periods (1999-2001, 2003-2004, 2005-

2007). Bribery Meanct and Bribery Dispersionct are the mean and standard deviation of the

frequency to pay bribes computed from BEEPS in cluster c. The coefficients of interest are

β1 and β2. Their positive signs would favor the ‘grease the wheels’ hypothesis of corruption.

The vector Xit−1 stands for the vector of firm-level control variables. They are

measured at the beginning of each time period (i.e. at 1999, 2002, and 2005) to control for

the initial conditions, to reduce possible endogeneity between them and firm performance

measures. The full set of control variables is described in the section 3. The term υi removes

unobserved firm fixed effects that can create across-time correlation of the residuals of a

given firm (e.g. managerial skill). The term νt removes unobserved time fixed effects that

can be responsible for the correlation of the residuals across different firms in a given year

(e.g. aggregate shocks or business cycles). The term ςs captures unobserved firm-size fixed

effects (micro, small, and medium-large firms) that can lead to the correlation of the

residuals across firms of a given size class due to, e.g., specific regulations attached to firms

of a particular size;16 and εit is the i.i.d. random component. We estimate specification (1)

using standard errors robust to heteroskedasticity and clustered at the firm level. In addition,

we account for influential observations using Cook’s distance as the data for CEE countries

are highly volatile.17

Finally, we are concerned about measurement error in the bribery variables. Under

the assumption of the classical measurement error – it does not correlate with the error from

a regression – the coefficients of interest would be biased towards zero. This assumption

16 We control for firm-size fixed effects, because firm size is included in the criteria defining clusters, andsome firms move from one size category to another over time. The country, location and industry factors fromthe criteria are removed when firm fixed effects are taken into account. The exclusion of firm-size fixedeffects, however, does not affect the results.17 Cook’s distance is a measure based on the difference between the regression parameter estimates andwhat they become if the ith data point is deleted . Observations, for which this distance exceeds 4/N areremoved as outliers, where N is the number of observations used in the regression (Cook, 1977).

15

seems plausible as we use combined two independent datasets. In addition, we believe that

the possible measurement error is averaged out in our bribery mean measure; this, however,

may not be a case for bribery dispersion. The retained measurement error, therefore, could

be a second source of an attenuation of the estimates.

5. Results and Discussion

5.1. Baseline Results

Table 3 reports the results of the estimation of specification (1). Odd columns present the

results for the dependent variable, real sales growth, and even columns – for labor

productivity growth. In columns I–IV we control for time, country, industry, location, and

firm-size fixed effects. In columns V–VIII we add firm fixed effects.

If better firm performance is generally associated with higher participation in

bribery, then in within-cluster regressions (columns I–IV) should be biased upward,

because cluster-level firm performance may more likely affect cluster-level bribery.

Controlling for firm fixed effects should reduce this bias. Indeed, is smaller in columns

V–VIII compared to columns I–IV, advocating for the lessening of endogeneity bias, and

the use of firm fixed effects regressions.18 Any further attempts to control for endogeneity

would reduce this bias in the same direction.

The comparison of columns I–II with III–IV and of columns V–VI with VII–VIII

also shows that the inclusion of bribery dispersion variable into regressions does not change

the sign or significance of . This permits us to analyze both bribery mean and dispersion

variables together. The bottom of Table 3 shows the average effects of the bribery mean and

dispersion on firm performance as well as their sum.19

18 In the case of bribery dispersion, in within-cluster regression, its impact on firm growth is alsomore diluted than in within-firm regression.19 We compute the average effects as a product of the sample average value of the bribery mean orbribery dispersion and the corresponding estimated coefficient. For example, in column VII ofTable 3, the average effect of the bribery mean on sales growth is (−0.096×0.311)×100% ≈ −2.97%,

16

Using the regressions from columns VII and VIII as benchmarks, ceteris paribus,

the increase in bribery mean by its average value is associated with a 3.0% and 4.3%

decrease in corresponding firm performance measures. These numbers are relatively large

since the average real sales growth is 4.6% and the average labor productivity growth is

negative 3.1% in our sample. The results thus show that bribery is a burden for firm

performance, which is consistent with some previous findings at both the micro (Fisman and

Svensson, 2007; De Rosa et al., 2015) and macro level (Mauro, 1995; Campos et al., 2010).

The estimates of the coefficients on bribery dispersion, in contrast, are positive for

both dependent variables, and highly significant. For a given level of bribery mean in a local

market, the average bribery dispersion effects are 4.7% and 5.9% for the two performance

measures. The sum of the average bribery mean and dispersion effects is positive and equals

1.7% for sales growth and 1.6% for labor productivity growth.

These results suggest that while higher level of bribery impairs sales and labor

productivity growth, firms grow faster in local environments with higher dispersion of

individual firm bribes. This suggests that bribery ‘greases the wheels’ of doing business for

individual firms, but harms firms’ collective economic performance. In more dispersed

environments, firms that are more efficient in bribery – those that have more information

about ‘greasing the wheels’, are discriminated by the public officials in a mutually

beneficial way, with lower costs or higher willingness to bribe – apparently bribe more

frequently. Owing to bribes, they most likely generate higher growth rates than if they were

not to bribe. Their non-bribing (or less frequently bribing) counterparts must be more

efficient in production and growth to compete with bribing firms. In this case, both types of

firms are able to generate increasing sales and labor productivity growth rates within the

local market.

where −0.096 is the estimate of the coefficient on the bribery mean and 0.311 is the sample averageof the bribery mean variable.

17

In less dispersed local bribery environments, if the number of bribing firms prevails,

negative externality from bribery (such as incentives to induce the bureaucratic burden by

public officials) can slow growth rates. If the number of non-bribing firms dominates, then

there can be fewer incentives for firms to be more efficient and compete aggressively with

occasionally bribing firms.

The results also show that the effects of bribery mean and dispersion are sounder for

labor productivity than for sales growth rates. This suggests that participation in bribery

affects the employment structure of firms. In highly corrupt environments, firms likely

employ a non-optimal (higher) number of workers due to misallocation of talent, in

accordance with Murphy et al. (1991) and Dal Bo and Rossi (2007). It may also be the case

that public officials (or local government), having established a connection with a firm, do

not allow it to dismiss its workers in order to keep high employment figures in the region

and therefore more loyal voters. However, bribing firms that have an opportunity to gain a

competitive edge over their non-bribing counterparts (in more heterogeneous environments)

are able to adjust the employment structure to an optimal level and increase effectiveness.

The results thus show that bribery can work as the ‘grease the wheels’ instrument,

despite its overall damaging effect. The existence of a certain number of bribing firms in a

local market increases aggregate firm performance as the positive effect from the bribery

dispersion exceeds the negative effect from the bribery mean. This is in line with Acemoglu

and Verdier (2000), Infante and Smirnova (2009) and De Vaal and Ebben (2011).

The following subsections examine the effect of bribery with respect to the

heterogeneity of firms and environments to understand better what drives the relation

between bribery and firm performance. The last subsection describes robustness checks.

5.2. Heterogeneity of Firms

5.2.1. Manufacturing and Service Firms

18

In our dataset, firms from manufacturing sectors represent only 14.5% of the sample. On

average, they tend to have lower sales growth, higher labor productivity growth, and pay

bribes less often than firms from service sectors.20 Panel A in Table 4 presents the results of

the estimation of specification (1) for manufacturing, service, and construction firms

separately. In this table, for the sake of space, we present only average bribery mean and

dispersion effects, while the full estimation results are available in the Online Appendix.

The estimated coefficients on bribery mean and dispersion are different for the

manufacturing and services firms.

Higher bribery mean in a local environment significantly reduces the performance of

manufacturing firms, especially the real sales growth. Operating in more bribery

heterogeneous environments does not bring them benefits either (see columns I–II, Panel A

in Table 4). One explanation of this result is that larger manufacturing firms are more

visible and attractive to corrupt public officials. At the same time size can make these firms

less flexible in responding to the bribery and lessen the capacity to extract benefits in

dispersed local bribery environments. Manufacturing firms also tend to have a larger share

of foreign ownership and exports, which is usually associated with higher management

standards, stricter attitudes against corruption and, perhaps, a poorer ability to deal with it.21

Another explanation for the result may be that our bribery measure does not reflect

well the nature of corruption practices among manufacturing firms. These firms arguably

require fewer permits, licenses, and inspections than do service firms but might depend

more heavily on relationships with customers and supply chains. Their corruption practices,

therefore, might instead consist of kickbacks between businesses.

Service firms, in contrast to manufacturing, are usually smaller, more flexible, and

likely to interact more often with public officials. Although on average they suffer as well

20 These statistics are available in the Online Appendix.21 Unfortunately, data limitations do not allow us to control for firm ownership structure or export shares.

19

from a higher bribery mean, they are also able to gain significantly in local markets with

higher bribery dispersion, see columns III-IV in Panel A, Table 4. To a large extent these

service firms belong to wholesale and retail industries as they represent about a half of the

whole sample. Approximately 15% of the sample belongs to the construction industry. The

last two columns in Panel A, Table 4 show that construction firms are able to gain very high

returns in dispersed bribery environments, possibly related to bribery associated with public

construction tenders, building permits, related regulations, etc.

5.2.2. Firm Size

The literature usually documents corruption as a greater obstacle for micro and small firms

than for large firms, and hence it impedes the performance of smaller firms more (e.g., Beck

et al., 2005 and Aterido et al., 2011). This is explained, for example, by the fact that smaller

firms have weaker bargaining power and less influence on public officials. In the present

study, however, bribery measures the frequency of paying bribes ‘to get things done’ and

may not reflect corruption as an obstacle. We observe that the bribery mean increases with

firm size, hence, we do not expect the same results as the cited literature suggests.

Panel B in Table 4 presents the results of the estimation of specification (1) for three

subsamples of micro, small, and medium and large firms. The signs of the coefficients on

bribery mean and dispersion are the same as in the case for the whole sample; the

magnitudes, however, are different for the three subsamples. It turns out that the growth

rates of micro firms are the least affected by bribery, larger firms suffer the most from

higher bribery mean, and small firms are able to extract the greatest benefits in more

heterogeneous local environments.

One explanation of this finding is that firms of different class size carry different

regulatory burden. These differences are usually designed to promote the growth and

development of small businesses and encourage entrepreneurship (WB, 2004; EC, 2011).

20

Smaller (micro) firms are often required to comply with softer regulatory standards such as

reporting and keeping records for inspections. They may also be exempted from some taxes,

or have lower tax rates (Gauthier and Goyette, 2014). Further, smaller amounts of bribes can

be extracted from firms with a smaller number of employees and turnover.

5.2.3. Firm Dynamics

The number of firms increases over time in our dataset, allowing us to capture firm

dynamics to some extent. Therefore, we examine whether local bribery environments affect

the performance of entering, exiting, and stable firms differently. About 8.5% of firms

remain in the sample during all three periods, 24.8% of the sample are new firms that appear

in the second period and remain in the third, and only 3.3% are those that have exited from

the sample in the last period. The remaining firms that are present in the sample only in one

time period, or only in the first and the third – are not considered. Due to the data structure,

the number of entering and exiting firms represents only a rough approximation of actual

firm dynamics.

Entering and exiting firms on average pay bribes more frequently and have lower

sales growth. Entering firms have negative and exiting firms have large positive labor

productivity growth rates, suggesting that the former are increasing and the latter are

decreasing the number of employees.22

Panel C in Table 4 reports the results of specification (1) estimated for these three

subsamples. The coefficients on the bribery mean and dispersion are significant and have

the same signs as in the estimates for the whole sample. However, bribery practices seem to

have a stronger effect on the performance of firms that are at the beginning or at the end of

their business experience. The strong negative impact of bribery mean on the growth rates

of exiting firms could be associated with costly bureaucratic exit procedures related to

22 See tables in the Online Appendix

21

bankruptcy or retreat from the market and final tax administration. These firms might also

attempt to fight for survival in the early stages of exit. Costly bribes for entering firms might

be needed for the firms to become established. It is notable that the sum of average bribery

mean and dispersion effects is negative for stable firms. This fact should incite incumbent

firms to protest against corruption.

5.3. Heterogeneity of Environments

5.3.1. Countries’ Institutional Environments

Although the countries from of CEE region underwent transition at approximately the same

time, they are heterogeneous with respect to the quality of formal and informal institutional

frameworks, as well as their overall corruption levels (Figures 1 and 2 in Appendix). Not

surprisingly, countries that entered the European Union in 2004 (Slovenia, Hungary, Poland,

the Czech Republic, Slovakia, Estonia, Latvia, and Lithuania) tend to have less corruption,

while Russia and Ukraine appear to be the most corrupt according to the Control of

Corruption indicator. In this section we analyze how local bribery environments affect firm

performance depending on the level of countries’ institutional strength.

We use the Rule of Law indicator from the WGI database to proxy for the strength

of countries’ institutions. It captures the incidence of crime, effectiveness of the judiciary,

and enforcement of contracts. We rescale this indicator to a variable that ranges from 0 to 1,

where higher values stand for a weaker Rule of Law. We augment the specification (1) with

interaction terms between the Rule of Law and bribery measures to see how country

institutions are associated with the bribery – performance relationship.

Columns I-II in Table 5 report the coefficients of interest from the estimation of

these specifications and Figure 3 in Appendix depicts the average bribery mean and

22

dispersion effects for different values of the Rule of Law indicator. 23 The results suggest

that although the bribery mean has a negative impact on firm growth rates, in countries with

weaker institutions this impact is less pronounced. In countries with the weakest Rule of

Law indicator, such as in Serbia between 1999 and 2001, the effect is even positive. The

weakening of institutions also decreases growth gains from the bribery dispersion in local

markets; they have, however, never become negative in our sample of countries. Hence, a

higher probability of being caught and stricter law enforcements make bribery more

detrimental, and the possibility to discriminate amongst firms brings higher benefits. We

thus provide empirical evidence for the theoretical conjectures of Infante and Smirnova

(2009) and De Vaal and Ebben (2011) that some amount of corruption can result in an

efficient outcome. We, however, contradict the empirical evidence of De Rosa et al. (2015)

showing that bribery is more harmful in non-EU countries.

5.3.2. Local bribery environments

Our baseline results show that, ceteris paribus, a higher bribery mean (dispersion) leads to

lower (higher) economic performance of firms. In this subsection, we examine the

interaction between these two characteristics of local bribery environments. In particular, we

are interested in how the bribery mean affects firm growth rates depending on the extent of

bribery dispersion.

Columns III-IV in Table 5 offer the results of estimation for the specification (1) in

which the interaction term between the bribery mean and dispersion is included. The

coefficient on the interaction term is positive, large and significant. This implies that firms

more likely increase their growth rates if they operate in local bribery environments, in

which corruption stakes and the opportunities to gain from corruption are higher. Figure 4 in

Appendix shows average bribery effects on sales and productivity growth rates for the range

23 Full results are available in the Online Appendix.

23

of bribery dispersion values. When bribery dispersion exceeds 0.4, these effects become

positive. This value of bribery dispersion falls into the 95th percentile of the sample

distribution. If bribery dispersion equals zero and bribery mean increases by its sample

mean value, then firms growth decreases to the negative rates -8% and -9% for sales and

labor productivity correspondingly. In the environments where all firms uniformly

participate in bribery practices, corruption is extremely harmful.

5.4. Robustness Checks

In this section we describe robustness checks which we conduct to verify the stability of our

results. The complete estimation results are available in the Online Appendix.

As a first robustness check we use bribery measures constructed as dummy variables

from the original frequency of paying bribes. The first measure takes value one if firms

report that they bribe public officials sometimes, frequently, usually, or always to ‘get

things done’, and zero otherwise, as in De Rosa et al. (2015). The second measure takes

value one if firms report that they bribe seldom, sometimes, frequently, usually, or always,

and zero if never. These measures are averaged within clusters. The bribery dispersion

variable is computed as before. The estimates of the coefficients of interest in the

regressions with new bribery measures remain similar to those from baseline results; only

their magnitudes are slightly smaller. The results therefore are not influenced by the

construction of bribery mean measure.

Although the bribery measure is consistently defined across three waves of BEEPS,

the structure of the questionnaire and stratification of surveyed firms were slightly changed

in the last wave. In addition, the number of firms in the Amadeus database increases over

time. To see if these changes impact the results, we estimate specification (1) separately for

the first and second, and for the second and third time periods. In addition, we estimate

specification (1) separately for Russia and Ukraine only, and for the rest of the countries

24

excluding Russia and Ukraine, since these are the most corrupt and represent about half of

the whole sample. Again, the estimates confirm that sample restrictions do not affect the

main outcome.

To ensure a stable structure of BEEPS within clusters, rather than unconditional

averaging of the bribery measure from BEEPS, we compute the bribery mean variable

keeping constant such firm characteristics as foreign ownership, export, and firm age. We

then use this ‘conditional’ bribery mean variable to estimate specification (1). The bribery

dispersion variable, meantime, remains the same. The main results stay qualitatively the

same. Structuring the BEEPS and Amadeus data within clusters in another way, we

compute the bribery mean and dispersion variables using the bribery measure from BEEPS

multiplied by the proportions of young and old firms within corresponding clusters from

Amadeus. When we use these weighted measures to estimate specification (1), the

coefficients of interest only increased in absolute values.

The main analysis assumes growth rates averaged over three years and control

variables measured at the beginnings of the three-year periods. As a robustness check we

estimate specification (1) on yearly data (nine years in total) with lagged control variables,

using two methods. First, we use conventional firm, firm-size and time-fixed effects

estimation as before. Second, we include a lag of the dependent variable among the

explanatory variables to control for autocorrelation in residuals and apply Arellano and

Bond’s (1991) dynamic panel data estimation technique.24 The coefficients of interest are

not qualitatively different from the main results, meaning that neither the data structure nor

possible autocorrelation drive the results.

In the main analysis we require each cluster to have no fewer than four observations

in order to compute bribery mean and dispersion. Obviously, the higher the number of

24 In particular, we estimate specification (1) in first differences and use the second lags of independentvariables (except for the bribery mean and dispersion, since they do not change across the three year periods)as instruments.

25

observations in a cluster, the better is the measurement accuracy of these variables.

Therefore, we also conducted the analysis under a constraint of no fewer than three

observations and no fewer than five observations in each cluster. The results are

qualitatively the same. The magnitudes of the coefficients of interest, however, become

larger when bribery mean and dispersion are computed more accurately.25

Given that we can only use location size, but not region in the criteria defining

clusters, we check if the results remain the same when location size is excluded from the

criteria. First, we estimate specification (1) for the subsample of firms located in the capitals

of countries, since they are the only cities exactly identified in BEEPS and Amadeus.

Second, we estimate specification (1) on the dataset when location size is omitted from the

criteria. Remarkably, the results remain qualitatively the same in both cases.

In our bribery mean variable, the measurement error and perception bias are likely

reduced due to averaging out. The aggregation, however, does not solve the problem of

missing data. About 10% of the sample in BEEPS does not report frequency of bribing; this,

however, is the smallest number relative to other questions about corruption. To check

whether missing values affect the results, we estimate specification (1) putting lower

weights on clusters with a higher number of missing observations. The weight is equal to

the ratio of the number of non-missing values to the total number of observations in a

cluster. The estimated coefficients of interest are nearly identical to those from the main

analysis, ruling out the problem of missing data in the original bribery measure.

For another robustness checks, we do not account for influential outliers using

Cook’s distance; we change the rule for defining outliers from 1% of top and bottom of

distribution to 5% and do not use data imputation (see Online Appendix). The estimates of

the coefficients of interest remain virtually the same as before and, therefore, robust to the

25 This fact also confirms that possible measurement error in our bribery variables could lead to attenuation ofthe estimates.

26

definition of outliers and the imputation procedure. The stricter rule for outliers accounting,

though, slightly increases the magnitudes of the coefficients on the bribery mean and

dispersion and doubles the overall fit of the regressions.

Finally, in specification (1) we add variables that measure different obstacles to

firms’ operation and growth obtained from BEEPS. These measures are averaged within our

clusters in the same way as the frequency to bribe. By including these obstacles we check

whether the bribery mean and dispersion explain the participation of firms in bribes, but not

other phenomena. We analyze the cases for corruption, tax administration, and obtaining

business licenses and permits obstacles. The inclusion of these obstacles into the

specification does not change the significance and signs of the main results.

6. Conclusion

This study empirically examines the relationship between ‘local bribery environments’ and

firm performance in Central and Eastern European countries. It provides an explanation for

divergent consequences of bureaucratic corruption found in previous research.

To overcome the data and methodological limitations of existing empirical literature

on bureaucratic corruption firm performance, we combine large and reliable firm financial

data from the Amadeus database with firm bribery practices data from BEEPS. We define

local markets by clusters of firms sharing the same country, industry, firm size and location

size characteristics. Within those clusters we compute the mean and standard deviation of

the frequency of paying bribes to public officials to ‘get things done’, and assign them to

each firm from the Amadeus database belonging to the same cluster. These two statistics

describe a local bribery environment: the equilibrium level of bureaucratic corruption in a

local market, and bribing behavior of firms shaped by firms’ willingness to bribe,

discretionary power of public officials to extract bribes, and uncertainty about

environments. The panel structure of the data, focus on local bribery environments, the use

27

of two independent data sources help us mitigate the endogeneity concerns between bribery

and firm performance measures.

Exploring within-firm variation, the results suggest that a higher bribery mean in a

local market retards both real sales and labor productivity growth. The increase in the

bribery mean by its average sample value is associated with about 3.0% and 4.3% decrease

in corresponding firm performance measures. This outcome complements some of the

existing empirical research on the consequences of corruption at the macro- and micro-

levels. We also find, however, that conditional on a given level of bureaucratic corruption,

higher bribery dispersion facilitates firm performance. The average bribery dispersion

effects are positive and equal to 4.7% and 5.9% for the two performance measures, so that

the trade-offs between bribery mean and dispersion are positive, too. These results are

robust to various specification checks and sample restrictions, and estimated at the lower

bound.

The results suggest that at least some bribing firms receive preferential treatments

from public officials, while non-bribing firms seem to be more efficient in production and

growth in more dispersed bribery environments. The presence of a certain number of

bribing firms in a local market increases aggregate performance, which is in line with

Acemoglu and Verdier (2000). High dispersion of individual firm bribes in some

environments can thus explain the persistence of corruption and advocate the ‘grease the

wheels’ hypothesis. In addition firms are more likely increase their growth rates in the

environments with higher both bribery mean and dispersion.

The main findings of the paper hold most strongly for services and construction

firms. The effects of a local bribery environment appear to be more important for firms with

more than 10 employees and for those that are at the beginning or at the end of their

business experience. The scope of our data however does not allow us to directly address

28

the impact of bribery on firm survival. The impact of bribery mean and dispersion on firm

performance also seems to be less sound in countries with weaker institutions.

29

References

Acemoglu, D. (2003). Root causes: A historical approach to assessing the role of institutionsin economic development, Finance and Development 40, 27–30.Acemoglu, D. and Verdier, T. (2000). The choice between market failure and corruption,The American Economic Review 90(1), 194–211.Anos-Casero, P. and Udomsaph, C. (2009). What drives firm productivity growth? TheWorld Bank Policy Research Working Paper 4841.Arellano, M. and Bond, S. (1991). Some tests of specification for panel data: Monte Carloevidence and an application to employment equations, Review of Economic Studies 58, 277– 297.Aterido, R., Hallward-Driemeier, M. and Pages, C. (2011). Big constraints to small firms’growth? Business environment and employment growth across firms, EconomicDevelopment and Cultural Change 59(3), 609 – 647.Beck, T., Demirguc-Kunt, A. and Maksimovic, V. (2005). Financial and legal constraints togrowth: Does firm size matter? The Journal of Finance 60(1), 137–177.Bliss, T. and Tella, R. D. (1997). Does competition kill corruption? Economic Policy105(5), 207–218.Boubakri, N., Cosset, J.-C. and Saffar, W. (2008). Political connections of newly privatizedfirms, Journal of Corporate Finance 14, 654–673.Campos, N., Dimova, R. and Saleh, A. (2010). Whither corruption? A quantitative survey ofthe literature on corruption and growth. IZA Discussion Paper No. 5334.Commander, S. and Svejnar, J. (2011). Business environment, exports, ownership, and firmperformance, The Review of Economics and Statistics 93(1), 309–337.Cook, R. D. (1977). Detection of influential observations in linear regression,Technometrics 19, 15–18.Dal Bo, E. and Rossi, M. A. (2007). Corruption and inefficiency: Theory and evidence fromelectric utilities, Journal of Public Economics, 9, 939-962.De Rosa, D., Gooroochurn, N. and Gorg, H. (2015). Corruption and productivity: Firm-levelevidence, Journal of Economics and Statistics 235(2), 115-138.De Vaal, A. and Ebben, W. (2011). Institutions and the relation between corruption andeconomic growth, Review of Development Economics 15(1), 108–123.Dollar, D., Hallward-Driemeier, M. and Mengistae, T. (2005). Investment climate and firmperformance in developing economies, Economic Development and Cultural Change 54(1),1–31.EC (2011). Minimizing regulatory burden for SMEs: Adapting EU regulation to the needsof micro-enterprises. Report from the Commission to the Council and the EuropeanParliament.Evans, D. S. (1987). Tests of alternative theories of firm growth, The Journal of PoliticalEconomy 95(4), 657–674.Fisman, R. and Svensson, J. (2007). Are corruption and taxation really harmful to growth?Firm level evidence, Journal of Development Economics 83(1), 63–75.Fungacova Z, Kochanova A. and L. Weill (2015). Does money buy credit? Firm-levelevidence on bribery and bank debt, World Development 68, 308-322.Gaviria, A. (2002). Assessing the effects of corruption and crime on firm performance:Evidence from Latin America, Emerging Markets Review 3(3), 245–268.Gauthier, B. and J. Goyette (2014). Taxation and corruption: Theory and firm-levelevidence from Uganda, Applied Economics, 46(23), 2755-2765.Hanousek, J., Kocenda, E. and Svejnar, J. (2007). Origin and concentration, The Economicsof Transition 15(1), 1–31.

30

Hanousek, J. and Palda, F. (2009). Is there a displacement deadweight loss from taxevasion? Estimates using firm surveys from the Czech Republic, Economic Change andRestructuring 42, 139–158.Huntington, S. P. (1968). Political Order in Changing Societies, New Haven: YaleUniversity Press.Infante, D. and Smirnova, J. (2009). Rent-seeking under a weak institutional environment,Economics Letters 104, 118–121.Jensen, N. M., Quan, L. and Rahman, A. (2010). Understanding corruption and firmresponses in cross-national firm-level surveys, Journal of International Business Studies41(9), 1481–1504.Kaufmann, D. and Wei, S.-J. (2000). Does “grease money” speed up the wheels ofcommerce? IMF Working Paper, WP/00/64.Klapper, L., Laeven, L. and Rajan, R. (2006). Entry regulation as a barrier toentrepreneurship, Journal of Financial Economics 82(3), 591–629.Lein, D.-H. D. (1986). A note on competitive bribery game, Economics Letters 22(4), 337–341.Lein, D.-H. D. (1990). Corruption and allocation efficiency, Journal of DevelopmentEconomics 33(1), 153–164.Lui, F. (1985). An equilibrium queuing model of bribery, Journal of Political Economy 93,760–781.Luo, Y. and Han, B. (2008). Graft and businesses in emerging economies: An ecologicalperspective, Journal of World Business 44(3), 225–237.Mauro, P. (1995). Corruption and growth, Quarterly Journal of Economics 110, 681–712.McArthur, J. and Teal, F. (2004). Corruption and firm performance in Africa, Developmentand Comp Systems 0409015, EconWPA.Meon, P.-G. and Weill, L. (2010). Is corruption an efficient grease? World Development38(3), 244–259.Murphy, K., Shleifer, A. and Vishny, R. (1991). The allocation of talent: Implications forgrowth, Quarterly Journal of Economics 106, 503–530.Rose-Ackerman, S. (1997). The political economy of corruption, in Elliot (ed.), Corruptionand the global economy, Washington DC: Institute for International Economics, 31–60.Shleifer, A. and Vishny, R. (1993). Corruption, Quarterly Journal of Economics 108(3),599–617.Svensson, J. (2003). Who must pay bribes and how much? Evidence from a cross section offirms, Quarterly Journal of Economics 118, 207–230.O’Toole C.M. and F. Tarp (2014). Corruption and inefficiency of capital investmentsin developing countries, Journal of International Development 26(5), 567-597.Vial, V. and Hanoteau, J. (2010). Corruption, manufacturing plant growth, and the Asianparadox: Indonesian evidence, World Development 38(5), 693–705.WB (2004). Review of small business activities. Washington, DC: World Bank.

31

Tables

Table 1: Summary statisticsMean Median SD Min Max

Sales growth 0.05 0.04 0.52 -3.92 4.53Productivity growth -0.03 -0.03 0.47 -3.58 4.55Total Assets 4.14 4.05 2.04 -4.14 14.75Employees 2.61 2.40 1.36 0.18 11.34Profitability 0.11 0.07 0.33 -5.96 33.71Market Share 0.00 0.00 0.02 0.00 1.00Leverage 0.67 0.67 0.47 0.00 24.83Cash Flow 0.06 0.03 0.36 -16.81 137.31Bribery Mean 0.30 0.29 0.14 0.00 0.80Bribery Dispersion 0.27 0.28 0.09 0.00 0.58Rule of Law 0.68 0.83 0.26 0.00 1.00

Note: The Table reports summary statistics of employed variables for the whole sample. Number ofobservations is 678381, non-missing for all variables.

Table 2: Pairwise correlations1 2 3 4 7 8 9 10 11 12

1 Sales growth2 Productivity growth 0.60 a

3 Total Assets -0.03 a 0.04 a

4 Employees -0.08 a 0.07 a 0.67 a

7 Profitability 0.04 a -0.02 a -0.06 a -0.02 a

8 Market Share -0.00 0.01 a 0.19 a 0.17 a -0.00 a

9 Leverage 0.01 a -0.02 a -0.12 a -0.11 a -0.25 a -0.03 a

10 Cash Flow 0.04 a -0.01 a -0.01 a -0.02 a 0.65 a -0.00 -0.25 a

11 Bribery Mean 0.00 -0.02 a -0.06 a 0.13 a 0.02 a -0.05 a 0.01 a -0.0012 Bribery Dispersion -0.00 -0.00 -0.16 a -0.03 a 0.02 a -0.05 a 0.00 a -0.01 a 0.57 a

13 Rule of Law -0.03 a -0.06 a -0.25 a 0.12 a 0.02 a -0.09 a 0.01 a -0.02 a 0.47 a 0.35 a

Note: The Table reports pairwise correlations between employed variables. Number of observations is 678381,non-missing for all variables. The symbols a, b, and c denote significance at the 1%, 5%, and 10% levels,respectively.

32

Table 3: Baseline results

(I) (II) (III) (IV) (V) (VI) (VII) (VIII)Sales Productivity Sales Productivity Sales Productivity Sales Productivity

Bribery Mean -0.039a -0.016a -0.057a -0.033a -0.042a -0.072a -0.096a -0.139a

(0.004) -0.004 (0.005) (0.004) (0.004) (0.004) (0.005) (0.005)Bribery Dispersion 0.056a 0.054a 0.174a 0.219a

(0.007) (0.006) (0.008) (0.007)Total Assets -0.019a 0.003a -0.019a 0.003a -0.075a -0.038a -0.076a -0.040a

(0.001) (0.001) (0.001) (0.001) (0.002) (0.002) (0.002) (0.002)Employees -0.260a 0.085a -0.260a 0.084a -0.072a -0.005c -0.070a -0.004

(0.002) (0.002) (0.002) (0.002) (0.003) (0.003) (0.003) (0.003)Total Assets Sq. 0.003a -0.002a 0.003a -0.002a 0.003a -0.002a 0.003a -0.002a

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Employees Sq. 0.017a -0.001a 0.017a -0.001a -0.009a 0.021a -0.009a 0.020a

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Profitability 0.009a -0.053a 0.010a -0.053a 0.003 -0.024a 0.001 -0.028a

(0.003) (0.003) (0.003) (0.003) (0.004) (0.004) (0.004) (0.004)Market Share -0.046a -0.406a -0.048a -0.407a -0.928a -1.108a -0.916a -1.045a

(0.012) (0.015) (0.012) (0.016) (0.081) (0.085) (0.078) (0.080)Leverage 0.042a 0.008a 0.042a 0.008a 0.032a 0.040a 0.031a 0.039a

(0.001) (0.001) (0.001) (0.001) (0.002) (0.002) (0.002) (0.002)Cash Flow 0.126a 0.047a 0.124a 0.048a 0.074a 0.031a 0.076a 0.037a

(0.004) (0.004) (0.004) (0.004) (0.005) (0.005) (0.005) (0.005)Time, Country,Industry, andLocation size FEs

yes yes yes yes - - - -

Firm, Time,and Firm size FEs - - - - yes yes yes yes

N observations 653,460 651,849 652,950 651,415 628,239 627,758 627,459 627,067N group 446,205 446,280 445,678 445,807R2 adjusted/within 0.081 0.074 0.081 0.074 0.218 0.111 0.224 0.117Average briberymean effect -1.22% -0.49% -1.79% -1.02% -1.29% -2.22% -2.97% -4.32%

Average briberydispersion effect 1.49% 1.45% 4.66% 5.87%

Average total effect -0.30% c 0.42% b 1.70% a 1.55% a

Note: The Table reports the results of the estimation of specification (1) for two performance measures asdependent variables: real sales growth and labor productivity growth. All control variables are measured at thebeginning of each time period (i.e., at 1999, 2002, or 2005). The average effects are the products of theestimated coefficients on bribery mean and dispersion and the sample average values of the correspondingvariables; the average total effect is the sum of these two effects. Standard errors shown in parentheses arerobust to heteroskedasticity and clustered at the firm level. Cook’s distance is used to account for influentialobservation. The symbols a, b, and c denote significance at the 1%, 5%, and 10% levels, respectively.

33

Table 4: Different types of firms

(I) (II) (III) (IV) (V) (VI)Sales Productivity Sales Productivity Sales Productivity

Panel A: Manufacturing and service firmsManufacturing Services Construction

N observations 88 917 88 960 442 567 441 964 96,137 96,402Average briberymean effect -6.99 a -3.95 a -1.08 a -3.57 a -3.74 a -5.05 a

Average briberydispersion effect -0.68 -2.37 a 3.65 a 6.30 a 10.42 a 7.79 a

Average total effect -7.67 a -6.32 a 2.57 a 2.73 a 6.68 a 2.74 a

Panel B: By firm size: Micro, small and large firms2–10 employees 11–49 employees 50+ employees

N observations 291 283 291 513 228 848 228 688 107 719 107 728Average briberymean effect -0.76 a -2.79 a -3.80 a -3.79 a -6.78 a -5.04 a

Average briberydispersion effect 0.87 c 2.83 a 7.55 a 7.20 a 4.11 a 3.83 a

Average total effect 0.11 0.04 3.74 a 3.41 a -2.67 a -1.21 a

Panel C: Stable, new-entrant, and exiting firmsStable New entrants Exiting

N observations 101 841 101 859 212 722 213 066 28 004 28 072Average briberymean effect -4.36 a -2.79 a -2.93 a -5.29 a -9.69 a -5.63 a

Average briberydispersion effect 4.01 a 2.60 a 5.74 a 6.07 a 6.20 a 11.73 a

Average total effect -0.35 -0.19 2.80 a 0.77 b -3.49 a 6.10 a

Note: The Table reports the average bribery mean and dispersion effects (in percentages) after estimation ofspecification (1) for different subsamples of firms for two performance measures as dependent variables: realsales growth and labor productivity growth. The average effects are the products of the estimated coefficientson bribery mean and dispersion and the sample average values of the corresponding variables; the averagetotal effect is the sum of these two effects. In Panel A firms are divided on subsamples of manufacturing (ISICcodes 15–36), services (ISIC codes 51–93) and construction (ISIC code 45) sectors. In panel B firms aredivided into subsamples of micro (2–10 employees), small (11–49 employees), and medium and large (morethan 50 employees) firms. In Panel C firms are divided into subsamples of stable (present in the sample duringall three time periods), new-entrant (present in the sample in the second and third periods), and exiting (presentin the sample in the first and second periods) firms. Standard errors shown in parentheses are robust toheteroskedasticity and clustered at the firm level. Cook’s distance is used to account for influentialobservation. The symbols a, b, and c denote significance at the 1%, 5%, and 10% levels, respectively.

34

Table 5: Heterogeneity of countries’ and local environments

(I) (II) (III) (IV)Sales Productivity Sales Productivity

Bribery Mean -0.274 a -0.438 a -0.271 a -0.291 a

(0.014) (0.015) (0.011) (0.011)Bribery Mean* 0.284 a 0.441 a

Rule of Law (0.020) (0.020)

Bribery Dispersion 0.312 a 0.284 a -0.040 a 0.033 b

(0.019) (0.020) (0.015) (0.015)Bribery Dispersion* -0.202 a -0.081 a

Rule of Law (0.027) (0.027)

Rule of Law 0.822 a 0.181 a

(0.020) (0.019)Bribery Mean* 0.688 a 0.588 a

Bribery Dispersion (0.040) (0.040)N observations 627,634 626,869 627,546 626,995N group 446,004 445,806 445,726 445,773R2 within 0.240 0.127 0.225 0.119

Note: The Table reports the results of the estimation of augmented specification (1) for two performancemeasures as dependent variables: real sales growth and labor productivity growth. In columns I-II, specification(1) includes bribery mean and dispersion variables interacted with the Rule of Law indicator. In columns III-IV,specification (1) includes interaction between bribery mean and dispersion variables. Standard errors shown inparentheses are robust to heteroskedasticity and clustered at the firm level. Cook’s distance is used to accountfor influential observation. The symbols a, b, and c denote significance at the 1%, 5%, and 10% levels,respectively.

35

Appendix: Figures

Figure 1: Control of Corruption indicator

Note: The Figure shows the variation of the Control of Corruption indicator across countries and timeperiods. For each time period the average value over three years is taken. Higher values stand for loweroverall corruption levels.

Figure 2: Bribery mean

Note: The Figure shows the variation of the bribery mean constructed from BEEPS (before linking it to firmfinancial data from Amadeus) across countries and time periods. Spikes represent 95 percent confidenceintervals. Higher values indicate a higher frequency of bribing.

-1.2

-0.8

-0.4

0.0

0.4

0.8

1.2

Slov

enia

Hun

gary

Pola

nd

Cze

ch R

ep

Slov

akia

Esto

nia

Latv

ia

Lith

uani

a

Bul

garia

Rom

ania

Cro

atia

Serb

ia

Rus

sia

Ukr

aine

1999-2001 2002-2004 2005-2007

0.0

0.1

0.2

0.3

0.4

0.5

Slov

enia

Hun

gary

Pola

nd

Cze

ch R

ep

Slov

akia

Esto

nia

Latv

ia

Lith

uani

a

Bul

garia

Rom

ania

Cro

atia

Serb

ia

Rus

sia

Ukr

aine

1999-2001 2002-2004 2005-2007

36

Figure 3: Average bribery mean and dispersion effects depending on countries’s institutions

Note: The Figure shows average bribery mean and dispersion effects (on the vertical axis, measured inpercent) depending on different values of the Rule of Law indicator (on the horisontal axis) for sales and laborproductivity growth rates. Higher values of Rule of Law correspond to stronger institutions.

Figure 5: Average bribery mean effects depending on bribery dispersion

Note: The Figure shows average bribery mean effects (on the vertical axis, measured in percent) depending ondifferent values of the bribery dispersion (on the horisontal axis) for sales and labor productivity growth rates.

-14

-12

-10

-8

-6

-4

-2

0

2

4

6

8

10

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Average bribery mean effect on sales growth

Average bribery mean effect on labor productivitygrowthAverage bribery dispersion effect on sales growth

Average bribery dispersion effect on laborproductivity growth

-10

-8

-6

-4

-2

0

2

4

0.06 0.12 0.17 0.23 0.29 0.35 0.40 0.46 0.52 0.58

Average bribery mean effect on sales growth

Average bribery mean effect on laborproductivity growth